Giving commands to a self-driving car: How to deal with uncertain situations?

Current technology for autonomous cars primarily focuses on getting the passenger from point A to B. Nevertheless, it has been shown that passengers are afraid of taking a ride in self-driving cars. One way alleviate this problem is by allowing give natural language commands car. However, car can misunderstand issued command or visual surroundings which could lead uncertain situations. It desirable detects these situations and interacts with solve them. This paper proposes model when given finds objects causing it. Optionally, question generated system describing included. We argue if explain human-like way, gain more confidence car’s abilities. Thus, we investigate how (1) detect their underlying causes, (2) generate clarifying questions passenger. When evaluating Talk2Car dataset, show proposed model, Uncertainty Resolving System (URS), improves 12.6% terms I o U . 5 compared not using URS. Furthermore, designed referring expression generator (REG) Attribute-Referring Expression Generator (A-REG) tailored setting yields relative improvement 6% METEOR 8% ROUGE-l state-of-the-art REG models, three times faster. • Detect uncertainty giving said uncertainty. Generate objects.